Abstract: Due to the existing pedestrian detection system can not meet the requirement of high detection rate, low false positive rate and fast detection, a pedestrian detection system based on fast cascade classification is proposed. This system consists of pretreatment and classification detection. In the classification detection phase, it selects parts of the optimal features by AdaBoost algorithm, fixes the false alarm rate of the training samples and designs a fast cascade classifier combining the advantage of the serial connected classifier. The single feature classifier uses the quick sort strategy to redesign which improves overall system performance. Simulation results show that the Area Under Receiver Operating Characteristic Curve(AUC), F-measure and G-mean results of the FastCascade are higher than traditional AdaBoost algorithm, UnderSampling algorithm and EasyEnsemble algorithm.

Key words: pedestrian detection, fast cascade, unbalance classification, feature selection, classifier

摘要： 行人检测系统难以同时具有高检测率、低误报率和较快的检测速度。为解决该问题，提出一种基于快速级联分类的行人检测系统。该系统包括预处理和分类检测2个部分，在分类检测阶段，利用AdaBoost算法选取部分最优的特征，通过固定训练样本的误报率，并结合串联分类器的优点，设计快速级联分类器(FastCascade)，其中，单特征分类器使用快速排序策略，以提高系统的整体性能。仿真结果表明，该FastCascade的接收者操作特征曲线下面积、F-measure和G-mean结果均高于传统的AdaBoost算法、UnderSampling算法和EasyEnsemble算法。

关键词: 行人检测, 快速级联, 不平衡分类, 特征选择, 分类器

CLC Number: 

• TP301